Send a task in one HTTP request — get the result as soon as it’s ready. Text, files, voice, external tools. Fully async, all through the API.
Live demo · paste your JWT and pick an assistant · the token stays in this tab only
Three steps between your code and an AI assistant’s answer.
POST /api/tasks with a JWT from Aist Backend.
Pick an assistant and pass the input — text, files or voice. You get the task’s uuid instantly.
Agent creates a chat and a turn in Aist Backend and hands the assistant’s configuration to Aist Core. If the assistant needs external tools, Agent calls them automatically.
Polling: GET /api/tasks/{uuid} — check the status until the task is done.
Webhook: pass a callback_url on creation — Agent notifies you the moment processing finishes.
Everything an assistant can do in the chat — from your code, through one API.
Tasks are queued and run in the background. Your server never holds a connection open — fetch the result by uuid or receive it via webhook.
Text, voice messages and files up to 1 GB in a single request: PDFs, spreadsheets, slides, archives, source code, audio and video. The assistant reads documents and spreadsheets on its own.
Declare functions from your own code on the task (client_tools). When the assistant needs your data, the task pauses and waits for the result — then picks up right where it left off.
Web search with sources, page reading, integrations configured on the assistant: it calls the HTTP APIs it needs and weaves the response into the result.
If the assistant has generators, images and clips are created right inside the task and arrive as files in output — no extra requests.
GET /api/assistants shows which assistants your token can use, and their models — which llm_model_id you can pick for a task.
Pass previous turns in turns — the assistant continues the conversation with full context, including files from earlier tasks.
Pass a callback_url — the result is posted to your URL as soon as it’s done, signed with HMAC-SHA256 using callback_secret and retried on failure.
A concurrent task limit, max_execution_time per task, cancelling paused tasks, a list of your tasks. Clear errors instead of raw server responses.
From a token to a working integration in minutes, not days.
Paste your JWT — see when the token expires, your limits, the available assistants and their models, and your tasks. The playground sends a task without writing code, and every assistant comes with a ready example in curl, JavaScript and Python.
Open the dashboard →
runTask polls the status and runs the client tool loop for you,
uploads files, lists assistants and limits. Node 18+ and the browser.
What’s new, what changed in the API and what got fixed — all in one place, latest first.
What’s new →Agent takes AI assistants beyond the chat window.
Connect Telegram, VK or any other messenger. The bot forwards the user’s message to Agent, polls the status and sends the answer back. Every assistant capability — web search, image generation, working with documents — is available right in the chat.
Triage incoming requests, classify documents, generate reports. Your backend queues the tasks, Agent runs them in parallel with the right assistant and returns a structured result.
Add AI features to your mobile or web app without heavy infrastructure. Agent orchestrates everything between your app, the assistant store and the AI engine.
Upload a batch of files or texts and run one assistant over all of them. The concurrent task limit keeps the system from overloading, and Horizon spreads the load across workers.
Your API key is a JWT for your Aist Ai account. Issue one yourself in a couple of minutes, or let us set things up together for your company.
At chat.aist-ai.com with email, phone, VK ID or Yandex ID. No account yet? Sign up right there.
“Настройки” (Settings) → “API-ключи” (API keys): name the key and press “Выпустить” (Issue). The key is shown only once — save it right away. It’s valid for a year and can be revoked in the same place.
You’ll see the available assistants, your limits and ready-to-run code for the first request in curl, JavaScript and Python.
The API gives you the same assistants as the app, including your own. Tasks are charged to your account balance just like chat messages.
More details — getting access and how the token works.
One task — two requests. Below is a minimal example; every endpoint, format and scenario is covered in the docs.
Webhooks, files, multi-turn, input formats and the response structure — in depth, with examples and a reference.
Create a task and get the result in two requests.
Text, files and voice in the input array.
Uploading via POST /api/files and download links.
Delivering the result to callback_url and the HMAC signature.
Continuing a conversation with the turns[] array.
The output, messages and metadata fields of the response.
Agent is the bridge between your code and the AIST ecosystem.
The API gateway. Accepts tasks, queues them, orchestrates processing and returns the result. Laravel 13 + Horizon.
The central backend. Stores users, assistants, chats, turns, models, documents, tools and balances.
The AI engine. Models from Anthropic, OpenAI, Google, xAI, DeepSeek, Perplexity, Yandex and more. Runs turns and tools.
Middleware for the chat UI. Streams answers in real time over SSE. Runs side by side with Agent.
REST API → Redis queue → Aist Backend → Aist Core → polling or webhook. For server-side integrations, bots and automation. Async, no UI.
Browser → Aist Chat → Aist Backend → Aist Core → SSE stream. For the user-facing chat UI. Real-time, with a UI.
Technical details for whoever builds the integration.
| Parameter | Value |
|---|---|
| Authentication | RS256 JWT from Aist Backend; claims sub, sanctum, exp, abilities |
| Token revocation | A 403 from Aist Backend blacklists the JWT automatically; subsequent requests return 401 |
| Task limit | Up to 5 active (pending, processing, interrupted) per user; above that — 429. Paused tasks can be cancelled |
| Execution time | max_execution_time in seconds per task, up to 24 h; once it runs out, the task closes itself |
| Acceptance timeout | If aist-core doesn’t confirm it has accepted the task within 30 s, the task moves to failed automatically |
| Task error codes |
Assistant is not available — the assistant isn’t available;Insufficient balance — not enough balance;Token revoked — the token was revoked;Context window is full … — the history in turns doesn’t fit the model
|
| Webhook delivery | Up to 5 attempts with backoff of 10 s, 30 s, 60 s, 120 s; headers X-Task-Uuid and X-Signature (if callback_secret is set) |
| File uploads | Any type, up to 1 GB |
| Cost | The turn’s money fields (metadata.cost and others) are always zero — costs aren’t disclosed through the API |
| Task retention | Completed, failed and cancelled tasks are kept for 30 days, then deleted |
Let an AI coding assistant do the integration. llms.txt holds the full API documentation in a format built for LLMs.
Claude Code, Cursor, Windsurf and other coding assistants can read this file and generate working integration code right away — with authentication, file uploads, polling and webhook verification.